How We Use AI

Last updated: 27 August 2026

How Handover uses artificial intelligence and how we protect your operational data.

How Handover uses AI

Handover uses OpenAI's API, not the consumer ChatGPT product, to power features such as:

Transcribing voice notes into text when transcription is processed by OpenAI.
Classifying voice and text captures into departments, priorities, and report sections.
Generating executive summaries for handover reports.
Analysing uploaded template documents to extract structure and content.
Extracting contact, credential, supplier, and procedure details from uploaded images, screenshots, and pasted text in the library.
Short AI briefings when you publish a handover to incoming crew.

What data is processed

Depending on the feature, this may include the raw audio of voice notes you record, transcribed voice notes, capture titles and descriptions, rotation and vessel context, text extracted from PDF or Word documents you upload for template scanning, pasted library import text, and uploaded images or screenshots used for contact or library imports.

Audio may be sent to OpenAI and/or Deepgram for speech-to-text transcription. Handover does not store the original audio after transcription. Data is sent from our servers to OpenAI over encrypted HTTPS. Where audio is processed by OpenAI, OpenAI's published API data controls for /v1/audio/transcriptions state that this endpoint is not used for training and that it has no abuse-monitoring retention or application-state retention.

Because transcription works on the spoken audio itself, the text-redaction setting cannot mask a voice recording before it is transcribed. In the current library image-import flows, images are used only to extract structured details and are not retained by Helmwise as standalone stored images after extraction.

Photos attached to ordinary captures are stored with the capture in Handover. Attaching a photo does not, by itself, send that photo to OpenAI. A stored photo is sent to an AI provider only when you use a feature that specifically analyses the image, such as an image-based import or an enabled WhatsApp image workflow.

WhatsApp Capture

WhatsApp Capture is an optional, opt-in feature on paid plans. If you connect a phone in Settings, notes you send to the Handover WhatsApp number become captures. Messages are carried to us by Twilio, our WhatsApp messaging provider. WhatsApp itself is operated by Meta, and your use of WhatsApp is governed by Meta's own terms.

Voice notes sent over WhatsApp are always transcribed by OpenAI (gpt-4o-transcribe) and the audio is discarded after transcription. Images sent over WhatsApp are read by OpenAI to extract a structured note. The original image is stored in Handover as an attachment to that note, visible only to crew who can see the note itself. Deleting the note removes access to the photo in Handover, but the underlying file may remain in protected storage. Because the redaction setting cannot mask audio or the contents of an image, a WhatsApp voice note or photo, for example a business card, is processed as sent. Disconnect a phone in Settings to stop capture from it.

Privacy protections

PII redaction is on by default. For each vessel you can enable "AI privacy protection" in Settings. When enabled, we perform best-effort redaction of common personally identifiable information patterns, such as names, phone numbers, email addresses, flight references, and similar patterns, on our servers before sending text to OpenAI. The model works with placeholder tokens. Sensitive fragments are mapped only in server memory and restored in the response you see in the app. Redaction is pattern-based and may not catch every identifier or sensitive detail.

API data policy. By default, OpenAI does not use data submitted via the API to train or improve its models unless the customer explicitly opts in. OpenAI may still retain limited abuse-monitoring logs under its API data policy. See OpenAI Enterprise & privacy for their current commitments.

Template scanning. Uploaded files are processed to extract text or structure for the requested workflow. Helmwise does not keep those files as ongoing application state in the product after analysis unless you separately save resulting content.

Image-based imports. When you choose image import for contacts, credentials, suppliers, or procedures, the uploaded image or screenshot is sent to OpenAI so visible details can be extracted. Because the text is embedded in the image itself, the text redaction toggle does not mask the image before processing. The same limitation applies to audio before transcription. In those import workflows, the image is processed transiently for extraction and is not retained by us as a standalone stored image after the request completes.

Your control

You can turn AI privacy protection off in Settings if you need to, for example if redaction interferes with a specific workflow. When it is off, unredacted text may be sent to OpenAI for processing. You will be asked to confirm before disabling.

If you do not want a document image or screenshot processed by AI, use manual entry instead of the image import flow.